rename directory

README.md

@@ -8,6 +8,6 @@ python run_pdqn.py
 
 ### Load trained model
 
-python load.py --algo='hppo' --load_dir='models/_HPPO' [--render --no_sync]
+python load.py --algo='hppo' --load_dir='models/HPPO_1v0' [--render --no_sync]
 
-python load.py --algo='pdqn' --load_dir='models/_PDQN' [--render --no_sync]
+python load.py --algo='pdqn' --load_dir='models/PDQN_1v0' [--render --no_sync]

envs/soccer_against_keeper.py

@@ -18,7 +18,7 @@ class SoccerAgainstKeeperEnv(SoccerScoreGoalEnv):
 
     """
 
-    def __init__(self, sync=True, offense_on_ball=True):
+    def __init__(self, sync=True, offense_on_ball=False):
         self.offense_on_ball = int(offense_on_ball)
         super(SoccerAgainstKeeperEnv, self).__init__(sync=sync)
         low0 = np.array([-1, -1, 0], dtype=np.float32)

run_pdqn.py

@@ -51,6 +51,7 @@ if __name__ == '__main__':
     ep_returns = []
     lens = []
     max_return = 0
+    total_steps = 0
     for ep in range(args.num_episode):
         state = env.reset()
         ep_r = 0
@@ -59,6 +60,7 @@ if __name__ == '__main__':
             q.append(state)
         stack_state = np.array(q).flatten()
         for t in itertools.count():
+            total_steps += 1
             hybrid_action, action, params = agent.choose_action(stack_state)
             next_state, reward, done, info = env.step(hybrid_action)
             q.append(next_state)
@@ -76,7 +78,7 @@ if __name__ == '__main__':
         if (ep + 1) % args.save_interval == 0:
             avg_return = np.mean(ep_returns[-args.save_interval:])
             writer.add_scalar('reward/episode', avg_return, ep + 1)
-            writer.add_scalar('length/episode', np.mean(lens[-args.save_interval:]), ep + 1)
+            writer.add_scalar('episode_length/episode', np.mean(lens[-args.save_interval:]), ep + 1)
             if avg_return > max_return:
                 agent.save_model(args.save_dir)
                 max_return = avg_return